1 Lessons from IPv6 Steven M. Bellovin

2 History ● Effort started in ● Original primary purpose: more IP addresses ● Several competing proposals

3 Context ● Internet security starting to be noticed – Note: security ≠ cryptography! ● No deployed autoconfiguration – DHCP dates to 1993 ● No NAT; first RFC was 1994 ● OSI vs TCP/IP wars ● Routing table size an issue

4 Basic Decisions ● Do not change basic semantic model of IP; keep it simple – Slightly modified over time... ● Support other things seen as necessary – Mobility – Renumbering – Multicast ● Add security: IPsec ● Do not change TCP

5 What Happened ● Near-stalemate in the IETF among different proposals – IPng area and directorate formed ● Arguments over cryptography (vs. U.S. export laws) ● Routing table size was seen as not a primary issue – Use CIDR and easy renumbering ● Engineering took a lot longer than expected

6 Second System Effect ● Many features were added to IPv6 ● Scoped addresses – Required changes to the socket API ● Neighbor Discovery replaced ARP – Includes basic autoconfiguration ● Flow labels (but usage wasn't specified) ● Early decision on stateless autoconfiguration froze part of the address format

7 The Claims for V6 ● Bigger addresses – True, but doesn't attract end users ● Autoconfiguration – We now have DHCP for client configuration ● More secure – IPsec exists for IPv4, too

8 Engineering is Hard ● Some features were much more complex than people thought – Neighbor Discovery couldn't be secured with IPsec – Site-local addresses interacted poorly with the DNS ● Finishing the design took a lot longer than expected ● Renumbering is easier, but still not easy; there are too many addresses in configuration files, access control lists, etc. ● Multihoming is still unsolved

9 What Happened? ● The opportunity attracted too many feature creatures ● During that time, IPv4 didn't stop evolving – Many of the putative advantages of IPv6 became part of IPv4 ● Tighter allocation controls by the RIRs helped ● NATs solved a large part of the address space crunch

10 The World Changed under IPv6 ● Mobility hasn't been that important at the IP layer – Layer 2 mobility often works well: access points, cell phones, etc ● Firewalls and VPNs have eroded much of the application base for mobility ● ISP-based multihoming has become increasingly important – and it was explicitly ignored in the IPv6 design

11 What are the Lessons? ● The devil is in the details – you generally don't know what will work, or how well it will work, until you've built it ● Understand what the real issues are – IPv6 got big addresses right; it didn't understand multihoming and looked too much at mobility ● Understand what will motivate users to adopt it – NAT made big addresses irrelevant for most end users